Olivier Sprangers

Results 11 issues of Olivier Sprangers

I am trying to understand the Tweedie [metric](https://github.com/microsoft/LightGBM/blob/b857ee10cc9a913e6dedd15c2475765d1e923c7b/src/metric/regression_metric.hpp#L300) and [objective](https://github.com/microsoft/LightGBM/blob/b857ee10cc9a913e6dedd15c2475765d1e923c7b/src/objective/regression_objective.hpp#L728). In my simple mind, for a given loss metric, the gradient used in the corresponding objective should be the derivative...

question

This PR reduces the amount of device transfers when creating tensors throughout the library.

Fixes #926 and fixes #924. - Fix of #926: Fixed Auto* models that do not support all required variables from BaseAuto Add unit test to all Auto* models to test...

Adds a warning that exogenous features in `df` are ignored if `X_df` is not provided. Fix for #294 until we support historical exogenous variables in TimeGPT.

fix

Similar to [this PR](https://github.com/Nixtla/hierarchicalforecast/pull/172), but: - Memory & compute efficient implementations of `ols`, `wls_struct`, `wls_var`. - Efficient implementation of computing P-matrix for all MinT methods. - Replaced eigenvalue check by...

Adds simple use case for evaluating different pricing scenarios when forecasting product demand for a set of products in retail.

documentation
feature

- Forgot to remove the old SDK reference link builder in docs, causing doc build failure on main (because the old category ID has been removed on readme.com)

fix

- Adds Colab flags to all notebooks - Adds Colab badge to capabilities notebooks

documentation
feature
fix

Feature: - Adds [MOMENT ](https://arxiv.org/pdf/2402.03885) to NeuralForecast, following [this implementation](https://github.com/moment-timeseries-foundation-model/moment/blob/main/moment/models/moment.py#L127). It is implemented (following the reference implementation) as a univariate model that does not support exogenous variables. Todo: - Testing...

feature

This is a large refactoring PR and open for discussion. The main goal of the PR is to unify API across different model types, and unify loss functions across different...